How do Microsoft, Alphabet, Amazon and Meta compare on AI-related capital spending? Start with the table below, which keeps each company’s reported period and spending definition visible. The figures cover company-wide infrastructure and do not isolate every AI dollar. Use the analysis that follows to connect investment with cash flow, margins and evidence of returns.
This comparison retains the original reporting periods for Alphabet, Microsoft, and Amazon, with Meta updated using its October 21, 2026 Q3 earnings release. It separates reported facts from interpretation and focuses on the question that matters to investors: whether rapidly expanding infrastructure can produce durable cash flows and returns above the cost of capital.
Reviewed by: Shingo Harada
Original review: September 12, 2026
Meta figures updated: October 22, 2026. Other companies’ figures and reporting periods are unchanged. Commitment disclosures in Meta’s October 22, 2026 Form 10-Q were reviewed October 23, 2026.
The comparison at a glance
| Company | Latest revenue | Latest operating cash flow | Latest capital-spending measure | Latest disclosed growth signal |
|---|---|---|---|---|
| Alphabet | Q2 2026: $119.8bn | Q2: $39.1bn | Q2 property and equipment purchases: $44.9bn | Google Cloud revenue +82%; Search & other +17% |
| Microsoft | FY2026: $331.8bn | FY: $182.9bn | FY cash additions to property and equipment: $115.9bn | Azure and other cloud services +41% for FY2026 |
| Amazon | Q2 2026: $200.6bn | Q2: $45.4bn | Q2 cash capital expenditures: $53.1bn | AWS revenue +37% |
| Meta | Q3 2026: $64.6bn | Q3: $32.8bn | Q3 CapEx including finance-lease principal: $34.9bn | Total revenue +26%; Family of Apps revenue $64.2bn |
The table is a map, not a league table. Alphabet and Amazon emphasize cash purchases. Microsoft reports both cash additions and a broader quarterly CapEx measure that can recognize the full value of a finance lease when it begins. Meta explicitly includes principal payments on finance leases. A defensible comparison must preserve those labels.
Want to check the spending yourself? Use our AI capital-expenditure framework to review revenue, cash flow and investment together.
Why the CapEx definitions matter
Capital expenditure is often treated as if it were a standardized line item. In practice, the cash-flow statement, supplemental earnings commentary, leases, vendor incentives, and asset sales can produce different numbers. The problem becomes material when quarterly spending reaches tens of billions of dollars.
- Alphabet: free cash flow is operating cash flow less purchases of property and equipment. Its Q2 2026 release reported $44.9 billion of such purchases.
- Microsoft: the cash-flow statement reported $115.9 billion of FY2026 additions to property and equipment. Quarterly management commentary described CapEx as $41 billion in Q4, including finance leases; cash paid for property and equipment was $35.8 billion.
- Amazon: cash capital expenditures include leasehold improvements and reflect incentives from property and equipment vendors and proceeds from asset sales. Q2 2026 cash CapEx was $53.1 billion.
- Meta: its highlighted CapEx measure includes purchases of property and equipment plus principal payments on finance leases. Q3 2026 was $34.906 billion, comprising $33.942 billion of purchases and $0.964 billion of finance-lease principal.
None of these figures is a pure “AI CapEx” number. Data centers, CPUs, GPUs, networking, cloud capacity, offices, fulfillment assets, and other infrastructure can sit inside a company-wide total. Amazon’s total also supports its fulfillment network; Alphabet’s supports Google Services, Cloud, and other operations. The proper language is therefore “AI-related infrastructure investment” unless the company has specifically identified a narrower amount.
Alphabet: a sharp acceleration funded by Search and Cloud
Alphabet’s 2025 revenue exceeded $400 billion, while full-year operating cash flow reached $164.7 billion and capital expenditures were $91.4 billion. Management said the vast majority of CapEx went to technical infrastructure, approximately 60% to servers and 40% to data centers and networking equipment. It initially expected 2026 CapEx of $175 billion to $185 billion.
On its July 22, 2026 Q2 earnings call, Alphabet raised its full-year 2026 CapEx outlook to $195 billion–$205 billion from the previous $180 billion–$190 billion range, primarily to bring capacity online faster to meet growing demand. This is company-wide guidance, not reported spending or a separately disclosed AI-only amount.
The pace had clearly accelerated by Q2 2026. Revenue was $119.8 billion, up 24% year over year. Google Cloud revenue rose 82% to $24.8 billion, led by enterprise AI infrastructure and solutions as well as core Google Cloud Platform services. Google Services revenue rose 15%; Search and other revenue increased 17%, and YouTube advertising grew 13%.
In the same quarter, operating cash flow was $39.1 billion and property and equipment purchases were $44.9 billion, producing negative quarterly free cash flow of $5.9 billion under Alphabet’s definition. Trailing-12-month operating cash flow remained a substantial $185.7 billion, against $132.4 billion of property and equipment purchases.
The economic case is broader than Cloud alone. Alphabet says AI supports model development, Cloud demand, user experience, and advertiser returns. That diversification may create several paths to monetization. The counterweight is that depreciation and data-center operating costs follow the investment. Investors should track whether Cloud profitability and advertising gains offset those costs rather than assuming revenue growth alone proves an adequate return.
For company-specific context, read our Alphabet stock analysis.
Microsoft: the strongest disclosed cloud monetization, with lease complexity
Microsoft’s fiscal year ends in June, so its FY2026 results overlap only half of calendar 2026. Full-year revenue was $331.8 billion and operating cash flow was $182.9 billion. Cash additions to property and equipment were $115.9 billion, up from $64.6 billion in FY2025.
Microsoft Cloud revenue reached $214.4 billion for FY2026, and Azure and other cloud services revenue increased 41%. In Q4, Azure grew 43% and Microsoft Cloud revenue was $59.3 billion, up 27%. These figures provide unusually visible evidence that large infrastructure investment is accompanied by strong demand. Microsoft disclosed that annual Azure revenue exceeded $100 billion in FY2026; its July 29 earnings release reports quarterly Azure performance as a growth rate rather than a separate quarterly revenue line.
Q4 CapEx was $41 billion in management’s broader presentation. About two-thirds was for shorter-lived assets, primarily CPUs and GPUs; total finance leases were $5.6 billion, while cash paid for property and equipment was $35.8 billion. Management expected the following quarter’s CapEx to exceed $50 billion. The distinction matters because finance-lease commencement can make the supplemental CapEx figure more volatile than cash payments.
Microsoft’s key risk is not lack of demand today but the durability and economics of that demand. Management said demand continued to exceed available capacity, yet Microsoft Cloud gross margin has been pressured by AI infrastructure and increased usage. Shorter-lived computing assets also require faster depreciation and replacement than buildings. Investors should compare revenue growth with gross-margin direction, cash conversion, and the useful lives assigned to equipment.
See our Microsoft stock analysis for the wider business and valuation framework.
Amazon: AWS acceleration, but company-wide spending is broader than AI
Amazon reported 2025 operating cash flow of $139.5 billion. Free cash flow fell to $11.2 billion because purchases of property and equipment, net of proceeds and incentives, rose sharply. AWS sales were $128.7 billion for 2025, up 20%, while AWS operating income reached $45.6 billion.
By Q2 2026, consolidated revenue was $200.6 billion and AWS growth accelerated to 37%, its fastest rate in 18 quarters, with an annualized revenue run rate of about $169 billion. Quarterly operating cash flow was $45.4 billion and cash capital expenditures were $53.1 billion. For the six months, cash CapEx reached $96.3 billion.
Amazon’s 10-Q says cash capital expenditures primarily reflected technology infrastructure—most of it supporting AWS growth—and additional fulfillment capacity. That wording prevents an analyst from assigning the full amount to AI. The same caution applies to management’s company-wide spending outlook, which also covers chips, robotics, low-earth-orbit satellites, stores, and logistics.
The clearest pressure point is free cash flow. For the 12 months ended June 2026, operating cash flow was $161.4 billion and property and equipment purchases net of proceeds and incentives were $169.0 billion, resulting in negative free cash flow of $7.6 billion. AWS growth and operating income are strong evidence of demand, but the investment test is whether future incremental cash generation exceeds this enlarged capital base.
Read our Amazon stock analysis for additional operating and valuation considerations.
Meta: AI investment serves advertising and a new model effort
Meta generated 2025 revenue of $201.0 billion and operating cash flow of $115.8 billion. Full-year CapEx, including finance-lease principal payments, was $72.2 billion. The company’s initial 2026 outlook was $115 billion to $135 billion, driven by investment in Meta Superintelligence Labs and the core business.
Q3 2026 revenue rose 26% to $64.601 billion, while CapEx including finance-lease principal reached $34.906 billion. Operating cash flow of $32.786 billion therefore produced negative free cash flow of $2.120 billion under Meta’s definition. For the first nine months, free cash flow remained positive at $11.049 billion. On October 21, Meta narrowed its full-year CapEx outlook to $135 billion–$145 billion from $130 billion–$145 billion. Family of Apps produced $64.158 billion of Q3 revenue, underscoring that advertising and engagement remain the current economic engine even as management pursues AI models, recommendations, ads, assistants, and new enterprise opportunities.
AI can improve ad ranking and content recommendations without appearing as a separate segment, so revenue attribution is inherently difficult. Q3 operating margin fell to 20% from 40% in the prior-year quarter, while total expenses rose 68%. The quarter included a $9.89 billion charge related to the August agreement with U.S. state attorneys general. Investors should not attribute the entire margin change to infrastructure; future depreciation, third-party cloud spending, and operating costs remain central monitoring items.
Meta’s October 22 Form 10-Q adds a different lens from quarterly CapEx: $454.52 billion of non-cancelable contractual commitments at September 30, including $121.41 billion due in 2027, and separately disclosed leases not yet commenced of $355.73 billion. These multi-period obligations include services and non-AI activities and should not be added to annual CapEx or ranked against other companies without equivalent disclosures and definitions.
October 26, 2026 funding note
As reviewed October 27, 2026, Meta’s October 26 pricing term sheet sets concurrent offerings totaling $37.5 billion before underwriter options: $25 billion of common-stock forward offerings at $625 per share and $12.5 billion of mandatory-convertible preferred depositary shares. Pricing and the initial conversion range are now set; the actual future conversion ratio remains market-dependent and subject to adjustments. Offering settlement is scheduled for October 29, 2026 and was not complete as of this review. Meta does not expect initial proceeds from the common shares sold by forward sellers; proceeds depend on later forward settlement, expected on or before November 19, 2027. A potential $60 billion Class C at-the-market program is separate and not agreed. Neither should be counted as completed funding or operating cash flow. Keep this financing development separate from the historical FCF table and the unchanged $135–145 billion 2026 CapEx outlook.
Meta also faces regulatory and legal risks related to its platforms. Those risks can affect the cash available for reinvestment and the returns earned from AI-enhanced advertising. See our Meta Platforms stock analysis.
How investors can judge whether the spending is working
1. Match investment with the closest revenue signal
For Microsoft and Amazon, Azure and AWS growth provide relatively direct demand signals. Alphabet combines a rapidly growing Cloud business with AI effects in Search and advertising. Meta’s AI benefits are mostly embedded in advertising performance and engagement. No single growth rate is perfectly comparable, but each company should show evidence that additional capacity is becoming productive.
2. Track operating cash flow and free cash flow over a cycle
A quarter of negative free cash flow can be rational when capacity is built ahead of contracted demand. Persistent deterioration is more concerning. Use a consistent company-specific definition and reconcile it to the cash-flow statement. Our free cash flow yield guide explains why a temporarily low yield can reflect either valuable reinvestment or an overextended capital program.
The customer-spending signal is only the starting point. Use the semiconductor cash-flow framework to test how foundries, memory suppliers, fabless designers, and IP licensors convert that demand into normalized free cash flow.
3. Measure returns, not just scale
More servers do not automatically create value. The long-run test is whether incremental after-tax operating profit supports an incremental return on invested capital above the company’s cost of capital. Apply the framework in ROIC vs. WACC, while allowing for the lag between construction, deployment, customer usage, and revenue recognition.
4. Watch depreciation and asset lives
Cash spending appears immediately, but depreciation reaches the income statement over time. Buildings can have long lives; accelerators and servers may become obsolete much faster. Changes to estimated useful lives can also alter reported profit without changing cash paid. Compare gross margins, depreciation, asset retirements, and impairment disclosures with the physical mix of spending.
5. Stress-test valuation assumptions
A valuation should not extrapolate peak infrastructure growth forever or assume margins recover automatically. Before building scenarios, use the Reverse DCF Calculator to identify the cash-flow growth the current price already requires. The DCF valuation guide then provides a disciplined structure for linking CapEx, depreciation, margins, reinvestment, and terminal value across a base case, a demand-upside case, and a case in which utilization, pricing, or technological obsolescence disappoints.
Capital spending does not become supplier value automatically. The AMD analysis follows the path from accelerator demand to margins and free cash flow; the Arm analysis tests how adoption becomes license and royalty revenue; and the SK hynix analysis examines whether HBM demand can produce attractive returns through a full memory cycle.
What the current evidence says
The 2026 disclosures show real demand rather than investment based solely on a distant promise. Azure, AWS, and Google Cloud all reported rapid growth, while Alphabet Search and Meta advertising continued expanding. Microsoft and Amazon also described demand that required additional capacity.
Yet the cash-flow cost is equally real. Alphabet produced negative quarterly free cash flow in Q2 under its definition; Amazon’s trailing-12-month free cash flow was negative; Microsoft’s property and equipment additions nearly doubled year over year; and Meta reported negative Q3 free cash flow of $2.120 billion and raised the lower end of its annual CapEx outlook to $135 billion. Revenue momentum answers the question “Is the capacity being used?” only partially. It does not yet settle “Will the lifetime return exceed the capital cost?”
A sensible conclusion is therefore conditional: Microsoft currently offers the clearest combination of large-scale cloud revenue and growth disclosure; Amazon and Alphabet show powerful acceleration with visible near-term free-cash-flow pressure; Meta is using infrastructure to reinforce an advertising franchise while funding a broader AI ambition. The relative investment case still depends on price, margins, capital intensity, and future returns—not on the largest CapEx headline.
Browse the AI Infrastructure Research Hub for company analyses and related comparisons.
Frequently asked questions
Which company is spending the most on AI?
The disclosures do not support a precise answer. Each company reports company-wide capital spending using a different definition, and each total includes assets that may serve AI and non-AI workloads. A ranking that labels every disclosed dollar as AI spending would overstate the evidence.
Does negative free cash flow mean the investment is failing?
No. Data centers are built before their full revenue arrives, so free cash flow can decline during a productive expansion. The warning sign would be sustained weak cash generation without corresponding growth, margins, contracted demand, or credible future capacity utilization.
What is the best metric for comparing AI infrastructure returns?
No single metric is sufficient. Combine cloud or advertising growth, operating cash flow, consistently defined CapEx, gross margins, depreciation, and incremental ROIC. Then test those operating assumptions in a valuation rather than treating CapEx itself as evidence of value creation.
Primary sources and methodology
- Alphabet, Q4 and full-year 2025 earnings call.
- Alphabet, SEC-filed Q2 2026 earnings release and Form 10-Q, plus the July 22, 2026 Q2 earnings-call transcript, page 13 (full-year CapEx guidance).
- Microsoft, FY2025 Q4 earnings release, FY2026 Q4 earnings release, and FY2026 Q4 call.
- Amazon, Q4 and full-year 2025 results, Q2 2026 results, and SEC-filed Q2 2026 Form 10-Q.
- Meta Platforms, Q4 and full-year 2025 results and SEC-filed Q2 2026 earnings release, plus the SEC-filed Q3 2026 earnings release, October 21, 2026, and Q3 2026 Form 10-Q, filed October 22, 2026 (commitment disclosures).
Figures are rounded from company filings. Growth rates are reported rates, not independently calculated estimates. Fiscal periods and capital-spending definitions are retained as disclosed; no attempt has been made to manufacture a standardized “AI-only” CapEx figure.
This article is for educational and informational purposes only. It is not investment, financial, tax, legal, or accounting advice, and it is not a recommendation to buy, sell, or hold any security. Company disclosures, capital plans, market conditions, and estimates can change. Verify all figures against the latest primary filings and consider your objectives and risk tolerance before making an investment decision.
For the supplier-side outcome of this spending cycle, compare how NVIDIA, AMD, and Broadcom convert AI infrastructure demand into margins, free cash flow, and returns on capital.